WBC Detection and Classification using Optimization Deep Learning – A Review

One of the main components of blood is white blood cells (WBCs). These cells provide protection against organisms that cause infections, such as bacteria, viruses, and fungi. White blood cells come in five different varieties in the human body. Several names for these cells exist, such as neutrophils, eosinophils, basophils, lymphocytes, and monocytes. There are several conditions that might arise from an overly high or insufficient white blood cell count. There are four different White Blood Cell (WBC) classifications in the dataset of 12,500 JPEG photos : Eosinophil, Lymphocyte, Monocyte, and Neutrophil. Each class has more than 3,000 images. By selecting a collection of these images and analyzing the region of interest (ROI), the approximate color of the WBC has been determined. Using clustering algorithms, the obtained color was utilized as a centroid for image segmentation. A new segmented dataset was created by cropping the ROI bounding boxes. The results were gathered and trained using the CNN ResNet 50 model. The dataset was split between test and train sets in an 80:20 ratio. Following 25 epochs, the model's learning rate was 1.0000e-06, its accuracy was 0.9998, its validation accuracy was 0.9654, its validation loss was 0.1378, and its loss was 6.2305e-04. WBCs and specific types can be detected and recognised by the system.